Bayesian estimation of discrete duration model

This thesis is comprised of three chapters which discuss Bayesian estimation of discrete time duration models.;The first chapter presents the multiperiod probit model and discusses estimation with a Gibbs sampler with data augmentation. As an empirical illustration, the multiperiod probit model is used to estimate a duration model using employment duration data for New Brunswick, Canada. The results from Bayesian estimation are compared with maximum likelihood estimation of a logit hazard model. Bayesian estimation of a model with unobserved heterogeneity is shown to be a simple extension of an estimation of a model with no unobserved heterogeneity.;The second chapter discusses parametric and nonparametric specifications of duration dependence. I then propose an alternative to these specifications that can capture the features of both specifications. I employ a Shiller smoothness prior to restrict the curvature of the parameters in the nonparametric duration dependence specification and hence put restrictions on the shape of the baseline hazard. The smoothness prior is used to help 'filter' the 'noise' that commonly appears in estimates of the baseline hazard. The methods are illustrated in an empirical exercise with employment duration data from the Canadian province of New Brunswick. The estimates with the smoothness prior are compared with the competing alternatives to modelling duration dependence.;In the third chapter a Bayesian estimator for a discrete time duration model is proposed which incorporates a nonparametric specification of the unobserved heterogeneity distribution through the use of a Dirichlet process prior. This estimator offers distinct advantages over the nonparametric maximum likelihood estimator (NPMLE) of this model. First it allows for exact final sample inference. Second, it is easily estimated and mixed with nonparametric specifications of the baseline hazard. An application of the model to employment duration data from the Canadian province of New Brunswick is provided.